Is It Time to Move Past HR/FB Rate?
Mike Podhozer put out the following question the other day asking if is luck or skill that Brandon McCarthy has such a high HR/FB%.
Prove that Brandon McCarthy‘s HR/FB Rate is Not Just Bad Luck
I started looking at the question several ways and came up with a final conclusion that HR/FB is probably not the perfect stat to use when trying to determine if a pitcher has been lucky or unlucky giving up home runs.
Let me start by going off on a tangent. I am of the camp that players with a huge upswing are the reason groundball pitchers, like McCarthy, have a higher than expected home per fly ball rates. All but the most upward swings will get on top of a sinking ball and drive the ball downward into the ground. The hitter with an upswing will be the ones hitting this sinking pitch. In my opinion, each pitcher will have a subset of players who swing in line with his pitch plain and crush those pitches for home runs.
For the preceeding reasons, I think it is tough to just look at all fly balls. Most fly balls are outs. It is those few that are really hit solid which need to be examined.
Well, I am going to use Inside Edge’s batted ball data to look at hard hit fly balls. Besides just having the fly ball classifications, they label a fly ball’s contact as weak, medium or well-hit. The value I am going to concentrate on is well-hit fly balls. Of all the home runs hit from 2012 to 2014, Well-hit fly balls account for 85.6% all home runs, well-hit line drives are another 13.7% and medium fly balls are 0.6%. The key for a pitcher to limit the home runs he gives up is to minimize the well-hit fly balls and line drives. Of all batted balls, well-hit fly balls happens 7.5% of the time and well-hit line drives 10.8% of the time.
Looking at McCarthy, here are his values for the two batted ball types
Season: WH FB%, WH LD%
2012: 6.0%, 13.5%
2013: 9.5%, 10.6%
2014: 4.8%, 12.5%
2015: 25.8%, 12.9%
Career: 6.1%, 12.5%
His 2015 well-hit fly ball values are not even close to his career norm.
Then, I looked to see if pitchers had an ability to prevent the hardest hit data. The answer is NO with the data I have so far. The Inside Edge fly ball data showed some signs of possibly reaching a stabilization point at some time in the future, but looking at year-to-year and two years to one year data no strong correlation exists. Here is a plot of the well-hit fly ball from season one to season two (min 200 batted balls).
The fly drive data shows a small signs of correlation, nothing close to a stabilization point. Here is a plot of the line drive data
Exactly zero correlation here.
These results don’t surprise me one bit. Looking at just FB/FB rate, it takes around 9.4 years for it to stabilize (1239 fly balls). A skill which stabilizes in nine years is probably not the best to use. I think we need to move on from HR/FB.
So getting back to the original question of if the HR/FB is luck or not. Truthfully, I would ignore it. In the big picture, we are trying to see if a pitcher has a “skill” of giving up home runs. With the data freely available to the public, I would start with his HR/9 rate. It is quick, but not perfect. Currently, McCarthy has a 4.5 HR/9 in 2015 and a career value of 1.1 HR/9. We would expect his 2015 to be closer to his career numbers and/or the league average than the inflated rate.
While Fan Graphs doesn’t have the exact stat, HR/Batted Ball is a probably a better stat to use to see if a player has an elevated HR rate. It can be found by dividing HR by all batted balls (LD+GB+BUH+FB). It takes just less than three years to stabilize, so it will tell use quicker if a pitcher will give up more or less home runs than the average pitcher. Here are McCarthy’s values over the years:
Season: HR per Batted Ball Rate
2005: 6.2%
2006: 6.7%
2007: 2.6%
2008: 4.1%
2009: 4.1%
2011: 2.0%
2012: 2.7%
2013: 2.7%
2014: 4.0%
2015: 19.4%
Career: 3.7%
The major league average over the time frame is 3.6%. McCarthy looks to be a little more home run prone, but not a whole lot more than the MLB average. So looking into the question, I can’t “Prove that Brandon McCarthy‘s HR/FB Rate is Not Just Bad Luck”. In my opinion, it is back luck. From now on though, I don’t think HR/FB should be used. Some pitchers just don’t give up fly balls, but more of those fly balls will go for home runs. If we want to find if a pitcher is being lucky or unlucky with home runs, we should use HR/Batted ball. Ground ball pitchers like McCarthy will have less fly balls, but the few batters who do square up on his pitches may give it a ride.
Well, the preceding is quite a bit to digest. Basically, a year’s worth of home run has been historically worthless. Even looking at more precise batted ball data, year-to-year correlations don’t exist. Instead of HR/FB or HR/9, I will probably begin using home run per batted ball to help determine “luck” when looking to see if a pitcher giving up more home runs than the league average.
Jeff, one of the authors of the fantasy baseball guide,The Process, writes for RotoGraphs, The Hardball Times, Rotowire, Baseball America, and BaseballHQ. He has been nominated for two SABR Analytics Research Award for Contemporary Analysis and won it in 2013 in tandem with Bill Petti. He has won four FSWA Awards including on for his Mining the News series. He's won Tout Wars three times, LABR twice, and got his first NFBC Main Event win in 2021. Follow him on Twitter @jeffwzimmerman.
Good stuff, Jeff. Could be useful to have HR / Batted Ball % on players pages.
Cool.
Good research here but I don’t entirely agree with your conclusion. The fact that seasonal HR/FB regresses almost all the way to the mean (once you adjust for a few other factors they help predict it, I have it taking ~1600 fly balls to stabilize) suggests that we should just look at FB% which regresses quickly and pay almost no attention to HR/9. Looking at HR/9 leads to the (IMO faulty) conclusion that McCarthy is HR prone whereas (even though he’s throwing more 4-seamers than he did in the past and should have a higher FB% than his career rate) his expected FB% suggests that he won’t be HR prone going forward.
I actually think HR/9 is not useful b/c it just leaves me wondering what the underlying elements (mostly HR/FB per FB%) are and I’d need to look at those to know what’s going on.
The problem is availability and the more I think of it, high K pitchers will inherently have a low HR/9 because of a lack of batted balls.
Would HR/OFFB% be more predictive than HR/contact%? Part of the reason flyball pitchers have lower HR/FB% is because more of their flyballs are popups.
And does it matter which individual stat we use? We should be past the point of looking solely at HR/FB% (or HR/contact%) to say a guy has been lucky/unlucky. Those rate stats should vary depending on whether a guy is a flyball or groundball pitcher. Isn’t the more important thing to compare a pitcher’s homeruns to what is expected for that type of pitcher?
SIERA already exists, and it varies the expected homerun rate based on the type of contact that pitcher allows.
Sorry, that should be “mostly HR/FB and FB%” NOT “mostly HR/FB per FB%” whatever that would mean.
Thanks for making this point. I was banging my head against the wall in the comments of the other article trying to make the point that arguing ground ball pitchers’ HR/FB was higher was not capturing his high overall HR total.
This has very little to do with HR/FB rate, but it was sparked by reading that it stabilized after 9 years.
Sometimes, I wish there was a place on fangraphs (rotographs, specifically) where people could math cowboy it up. Rotographs (and fantasy as a whole) by its nature, is basically a gambling on percentages website.
So, while “Good Math” is important, I think there’s a place for math that’s a little bit sloppy. Basically, what I’m saying is that we’re left with HR/FB stabilizing after 9 years, and I’d imagine that you’re basically waiting for it to cross the 50/50 mark, or the 75/25 mark. Sometimes though, it’s fun to live on the edge: Why not run the numbers and allow for 75% noise/regression? I’ve certainly gambled on less.
I know it’s *terrible* math, and people are liable to draw *terrible* conclusions from it, but as long as it’s provided within the context of being terrible math, I think it’d be fun. We don’t always need 95% confidence, or a miniscule p-value, sometimes it’s fun to say, “Brandon McCarthy, with an n of 25, basically has a 10% chance of being worse than league average” vs. “We can’t judge Brandon McCarthy for another 7 years and it’s just smarter to regress it almost wholly to the mean”
All this DFS stuff that EVERY SINGLE BLOODY FANTASY WEBSITE is pushing is basically predicated on bad math, so why not embrace the bad math movement?
This first step is to hire a chimp for all data entry cuz bad math really needs random data input as well.
I understand your point. If given the data and time, why would we ever ignore any data that had a reachable regression point? I understand the time issue it would take to sync all of the data for each player’s regression rates. But we could go by quarters or something like that.
I’ve been down this road before. I used data from starting pitchers only from 2004 through 2013 and (I think) at least 200 TBF.
Year-on-year Pearson correlation for logit(HR/(HR+OFFB*+IFFB)) was 0.1731. (*OFFB did not include HRs by default.)
Removing IFFB from the denominator got me an r of 0.1713.
Know what the best denominator was? HR+OFFB+IFFB+LD. Brought it all the way down to 0.1525. Ultimately, not that much of an improvement over your standard HR/FB.
For what it’s worth though, when I used HR+OFFB+IFFB+LD+GB+BU as the denominator it brought the correlation up to 0.3076, meaning HR/batted is not as “random” as HR/FB.
I should clarify that by “best”, I meant the least year-on-year correlation because I was trying to regress HRs out of the equation.
If you want to know what the “best” home run denominator was in terms of higher Pearson r, it’s HR+SO. logit(HR/(HR+SO)) had a year-on-year correlation of 0.5555. But I’m sure most people wouldn’t want to use that because HRs are not a natural subset of SOs as they are with FBs.
So back to Mccarthy, given that his HR/battted ball rate was only marginally worse than league average last year, how do we explain his inflated ERA from last year? Just the high BABIP?
i.e. even if his HR/9 regresses back to last year’s levels he still needs to fix something else too to get his ERA to line up with his SIERA…
Has anyone ever done a study of HR/FB percentages per pitch type? It seems like individual pitchers should have an xHR/FB based on their specific pitch mixes. An additional layer would be to work in park factors to create a true xHR/FB per pitcher. From that you might find that we should expect a higher HR/FB for McCarthy, though the current rate is unlucky.
I haven’t seen one and I about about went down this path, but had to reign myself in.
As I showed in the comments of the linked column, pitchers with a high HR/FB% rate throw a larger than normal percentage of pitches in the bottom third of the zone whether those pitches have sinking action or not, while pitchers with a low HR/FB% throw a larger than normal percentage of pitches up in the strike zone. But sure, HR/FB% is of limited usefulness given that it’s the total number of home runs that really matters.
I keep thinking that if a pitcher sets out to work down in the zone nearly exclusively then almost every pitch up in the zone is by definition a mistake. What percentage of home runs are on mistakes? (No research was involved in this post.)
We could probably figure this out with pitchf/x. Take an average (or median) pitcher by DIPS (not WAR) with several years of stable performance and look at where he throws 68% of his strikes. Then look at 68% of his balls. You can draw a grid based on the density of outs, quality of contact versus pitch type – I suspect that for a groundball pitcher, you would expect to see more HR as balls are thrown higher in the zone, so the vertical limit should be easy.
But, you’d also suspect the “true zone” would be deeper and wider, since balls that are outside the official strike zone are going to be intentional.
McCarthy probably doesn’t give up any homers when he gets the 2 weeks between starts he likes. When he has to take his regular 5th day turn, he gives up a few..
—
HR/FB% is fine data as is FB% and HR/9inn. What’s uncool is the regularization of the HR/FB% stat and then tacking it on to FIP to make xFIP.
Crazymaking – and in the end, xFIP really just turns the useful and original FIP stat into something that resembles very closely the strikeout ‘K’ stat.
Well that’s not true at all. xFIP is an improvement over FIP the same way that FIP was an improvement over ERA, as the rate at which you give up home runs is heavily influenced by luck. But just as we learned that BABIP isn’t completely independent of the pitcher, HR/FB% isn’t completely independent of the pitcher either.
jdbolick:
You are raising up these stats on a holy grail pedestal and that is wrong. Please go re-read the tittle of the article.
No, “we” didn’t eventually “learn that BABIP isn’t completely independent of the pitcher, HR/FB% isn’t completely independent of the pitcher either.”
This is the mindset of believing these fun baseball stats that are mainly used to describe/differentiate male professional baseball players are in fact some sort of scientific proof.
No, this isn’t a point in time on the baseball science learning curve – that’s just baseball stat geek flat earth snobbery (Not calling jd a snob, you get the idea). Samermetrics, I call it.
None of this is science – and someone trying to “take the noise out of home runs allowed by regularizing HR/FB%” will be regularly defied by the stats of the better pitchers.
Yes, of course the players affect the play – everyone reading this right now would have an MLB BAPIP of .000, and that’s not luck.
Buttle, not Tittle
If you want evidence for what jdbolic said about xFIP being a good predictor of future ERA: http://www.hardballtimes.com/fip-in-context/
Just so I’m clear: Inducing in-field flyballs is a skill, but once those flyballs leave the infield they become under the complete control of chaos theory?
I don’t understand why this actually useful information is being hidden in the RotoGraphs section. 😛
Anyway, it mostly seems to validate the idea of SIERA.
It’s hard to take seriously any study on this topic that completely ignores park effects.
Great analysis, thank you!
Pizza Cutter, Tom Tango an others have found the stabilization point for HR/FB long time ago. It is about the same amout of time as BABIP. I have seen anything from 7 seasons to 10 seasons. Should we ignore it? Depends on what the question is. If I want to project Carlos Martinez ERA for his first full season of a starter…of course I don’t need to know his HR/FB. If I am trying to look at Tom Glavine’s career, of course I want to know it. He put up monumental HR/FB numbers for 15 years and pitched so far past stabilization point that regressing his career rates to league average would be plain nonsensical. Want to know Clayton Kershaw’s Hr/FB skill? Well, regress appropriatly. Not all or nothing.
What are you trying to find out?
Trying to find out re:’McCarthy? Go ahead and ignore it now. But at some point, if he pitches long enough, you won’t be able to ignore it. It may be close to league average, but at one point, it is. I longer ignorable.
Salazar just got called up and is supposedly here to stay…
– Salazar or McCarthy?
A bit late to the party here, but I did find that heavy groundball pitcher generally give up harder contact on fly balls. McCarthy has been a slightly GB-heavy pitcher over the past 4 years, but not too extreme. So we might expect him to give up slightly more homers and get fewer pop-ups per fly ball, but not way worse than average (based solely on his GB/FB tendencies, at least).
For an individual pitcher over a small sample, xFIP is still better than FIP/ERA, just remember that it’s going to slightly favor GB pitchers. SIERA should be best since it takes more factors into account. Over a larger sample (multiple seasons or more pitchers) FIP is probably better than xFIP.
http://www.hardballtimes.com/are-groundball-pitchers-overrated/